Skill UI

This skill provides comprehensive best practices and Python code examples for optimizing API calls to Large Language Models (LLMs) via OpenRouter. It covers benchmarking model latency, implementing streaming to achieve low Time-To-First-Token (TTFT), and processing requests concurrently using asynchronous programming (asyncio). Use this when developing real-time, high-throughput applications that require predictable and fast model responses.

A comprehensive guide on handling API rate limits and throttling errors (429) when building high-throughput systems, particularly for LLM APIs like OpenRouter. This skill covers best practices, including using exponential backoff, implementing custom client-side rate limiters (Token Bucket), and managing concurrent requests for stable, reliable performance.

This skill provides comprehensive reference architectures for building production-grade AI applications using OpenRouter as a unified LLM gateway. It covers three scaling patterns: simple single-service, standard microservice (with caching and budget enforcement), and enterprise event-driven systems (with message queues and workers). Use this when planning system design, reviewing complex AI pipelines, or ensuring scalability and cost control in your LLM integration.

This skill implements a sophisticated, config-driven rules engine for OpenRouter API calls. It dynamically determines the optimal large language model (LLM) by evaluating a comprehensive context, including user subscription tier, remaining cost budget, specific task type (code, chat), required tools, and latency requirements. This is essential for building production-grade, cost-aware, and robust LLM orchestration systems.

This comprehensive guide provides developers with step-by-step instructions for migrating applications from direct calls to proprietary Large Language Model (LLM) APIs (such as OpenAI or Anthropic) to utilizing the unified OpenRouter gateway. It details essential code changes, including updating base URLs, managing API keys, adding custom headers, and correctly prefixing model identifiers. This ensures a seamless transition, enabling multi-model compatibility and simplified API management.

Optimizes prompts for LLMs to cut token usage, lower costs, and boost response quality by analyzing, condensing, and rewriting instructions; useful when teams need faster, cheaper, clearer AI conversations.

Outlines guarantees valid JSON/XML/code outputs by applying CFG/FSM-constrained token filtering, Pydantic schemas, and grammar-level sampling while supporting local Transformers, llama.cpp, and vLLM backends for fast inference.

Covers HuggingFace PEFT workflows for LoRA/QLoRA-based LLM fine-tuning, enabling sub-1% parameter updates on limited GPUs plus adapter serving tips.

This guide details the process of migrating search and LLM functionalities from legacy APIs (Google CSE, Bing, SerpAPI) to Perplexity Sonar. It demonstrates how to utilize the Strangler Fig pattern to build a robust adapter layer. The primary benefit is consolidating a complex, multi-step pipeline (Search -> Raw Results -> LLM Summarization) into a single, efficient API call, ensuring automatic citation handling and streamlined development.

This guide outlines the comprehensive reference architecture for building advanced, AI-powered search and research applications, modeled after Perplexity. It covers three core tiers: a searchable service with model routing and caching, a citation pipeline for validation and rendering, and an orchestrator for multi-query research synthesis. Use this when designing complex, production-grade, search-augmented LLM integrations or reviewing advanced system patterns.

This skill provides comprehensive patterns for building highly reliable applications that interact with external Large Language Model (LLM) APIs, such as Perplexity Sonar. It implements critical reliability techniques including Circuit Breakers, Model Fallback, Streaming Timeout handling, and Caching layers. Use this to ensure your application maintains high availability and graceful degradation even when external services experience intermittent failures or variable performance.

This skill enables the development of robust, event-driven pipelines interacting with the Perplexity Sonar API. Since Perplexity lacks native webhooks, this system simulates event patterns using advanced techniques: Streaming Server-Sent Events (SSE) for real-time data, job queues (like BullMQ/SQS) for asynchronous batch research, and cron-triggered monitoring for scheduled news and trend alerts. It provides patterns for building complex, reliable, and scalable AI applications.